NVIDIA-AI-Blueprints/video-search-and-summarization

Leverages NVIDIA NIM microservices (Vision Language Models like Cosmos Nemotron and LLMs like Llama Nemotron) orchestrated through the Model Context Protocol to provide unified tool access for VLM-based Q&A, semantic video embeddings, long-form summarization, and real-time anomaly detection. The architecture separates real-time video intelligence (extracting visual features and embeddings via microservices), downstream analytics (enriching metadata streams into actionable alerts), and agentic workflows that coordinate across video retrieval, perception, and language understanding components. Supports multiple deployment topologies from local Docker Compose to cloud instances with validated GPU configurations, targeting both video analysts needing 1-click setup and ML engineers requiring custom pipeline modifications.

432 stars.